an empirical examination of
TRANSCRIPT
Behavioral Research In AccountingVolume 14, 2002Printed in USA
An Empirical Examination ofCompeting Theories to Explain
the Framing Effect inAccounting-Related Decisions
C. Janie ChangSan Jose State University
Sin-Hui YenTamkang University
Rong-Ruey DuhNational Taiwan University
��������
������������������� ��������������������������������������������������������� ����������������� ������������������������������������������������ �������������������������������������������� ���������������������� ����������������������������������� ���� ����� ������� ������ ��������������������������������!����������������� ���������������������������������������"���� �������������������������#�������������� ������$������������ ���������������������� ������������������������������������������ ����������������������� ���������� ���������������������������%���������!�������������"���� ���������������� ����� ���� �����������������������������������������������������������������������������������&������ �� ���������!���� ����� ������� ���������������� ��������������������#������������������%��� �����������������������������������!� ��� �����������������������������������������'������������������������������� ���������������������������������(��������������������������������������������������������!������������������������ ���������)����
INTRODUCTIONMost accounting tasks require that accountants make judgments
in collecting and providing information for managers to use in decision
We express our thanks to the editor, Susan Haka, and two anonymous reviewers fortheir comments on earlier versions of this paper. Data from this study may be obtainedfrom the first author upon request.
36 Chang, Yen, and Duh
making. It is possible that managers judge or justify their decisionsbased on the manner in which accounting information is provided,while not paying adequate attention to the content of that informa-tion. Managerial decisions that result from such biases can have ad-verse consequences for corporations and their stakeholders (Ashtonand Ashton 1995). Thus, identifying the impact of the manner in whichdecision-related information is provided by accountants to decisionmakers is a critical step in understanding how accounting informa-tion should be collected and provided to maximize firm value.
Decision-making literature has shown that individuals responddifferently to the same decision problem if the problem is presentedin a different format (for reviews, see Kühberger 1998; Levin et al.1998). This phenomenon is referred to as a framing effect (Kahnemanand Tversky 1979; Tversky and Kahneman 1981). Tversky andKahneman used prospect theory to explain framing effects. This theorysupports the findings of many prior accounting studies in framing.However, inconsistent results documented in some recent psychol-ogy literature (see Schneider [1992] for a review) have inspired be-havioral researchers to address the limitations of using prospecttheory to explain framing effects. For example, Levin et al. (1998)demonstrated three types of framing effects: standard risky choice,attribute framing, and goal framing.1 Each type of framing effect rep-resents divergent mechanisms and consequences in decision mak-ing. Levin et al. (1998) indicated that prospect theory probably bestexplains the risky choice framing effect, but not the other two typesof framing effects. Even so, Levin et al. (1998) cast doubt on theability of one single theory (i.e., prospect theory) to interpret the em-pirical evidence of risky choices in different contexts. As Wang andJohnston (1995) indicated, framing effects on individual behaviorare a context-dependent choice phenomenon—not a general, cross-domain phenomenon. Converging evidence demonstrates that a fram-ing effect depends on task, content, and context variables inherentin the choice problem (Fagley and Miller 1997; Wang 1996).
1 The framing effect associated with the classic Asian disease problem (Tversky andKahneman 1981) is a typical example of a standard risky-choice framing effect. Inmanagerial accounting literature, most framing studies fall into this category. Inattribute framing, a single attribute of an event (e.g., the evaluation of a client’sinternal control system) is framed either positively (e.g., the strength of internalcontrols) or negatively (e.g., the risk of internal controls). The framing effect is as-sessed by comparing the judgments of the likelihood of framed events’ occurrence.In most auditing framing research, auditors are asked to indicate the likelihood ofan event, such as the reliability of a client’s internal controls. In this scenario,auditors often commit an attribute-framing bias. A goal-framing effect impacts thepersuasiveness of a communication. Very few accounting studies in framing be-long to this type.
An Empirical Examination of Framing Effects 37
Alternative theories for analyzing framing effects were developedin the early 1990s (Frisch 1993; Reyna and Brainerd 1991a, 1991b;Schneider 1992; van Schie and van der Pligt 1995). In particular, proba-bilistic mental models (PMM) and fuzzy-trace theory (FTT) have beenused to test framing effects in making standard risky choices(Kühberger 1995; Reyna and Brainerd 1991a, 1991b; Reyna and Ellis1994). Since these theories have been investigated only innonaccounting settings, whether these theories maintain descriptivevalidity in accounting contexts remains unknown (Fagley and Miller1997; Schneider 1992; Wang 1996; Wang and Johnston 1995).
Previous accounting research has adopted, explicitly or implicitly,prospect theory to address the framing effect phenomenon in auditing(Emby 1994; Emby and Finley 1997; O’Clock and Devine 1995), tax(Elffers and Hessing 1997; Martinez-Vazquez et al. 1992; Newberry etal. 1993; Schadewald 1989; Schepanski and Kelsey 1990; Schepanskiand Shearer 1995; Yaniv 2000) and managerial accounting contexts(Barton et al. 1989; Chow et al. 1997; Rutledge and Harrell 1993,1994; Sharp and Salter 1997). Almost all auditing studies have exam-ined the attribute framing effect, and most managerial accounting andtax research have investigated framing effects focusing on risky choices.Although most studies have confirmed prospect theory, the results ofJackson and Spicer (1986) and Martinez-Vazquez et al. (1992) did notsupport the proposition of prospect theory for income tax compliance.Moreover, Barton et al. (1989), using employees with managerial expe-rience, found atypical framing effects that were opposite to the predic-tion of prospect theory. Boritz (1997) questioned the appropriateness ofusing prospect theory to explain the framing effect in auditing contexts.
In tax studies, both Sanders and Wyndelts (1989) and Schadewald(1989) found no framing effect, but observed a reflection effect2 inmaking tax decisions. Most prior research has not distinguished fram-ing effects from reflection effects. Arkes (1991), Kühberger (1995), andLevin et al. (1998) argued that these two effects are different—the fram-ing effect concerns presenting the same decision problem with differentframes, while the reflection effect involves two different decision prob-lems. Failure to distinguish between these two effects may lead to mis-interpretation of some studies’ results or may make it difficult to explainresults with mixed signals. Moreover, because the framing effect is abias while the reflection effect is not, accounting professionals shouldbe alert to possible framing effects and design a reporting system toprevent such effects; they can be less vigilant about reflection effects.However, decision makers should be reminded by some means that
2 More detailed discussions of how to distinguish between framing and reflection ef-fects are provided later in this paper.
38 Chang, Yen, and Duh
reflection effects are related to two decision problems and that theyneed to consider factors in each situation separately.
In this paper, we compare the abilities of three theories (prospecttheory, probabilistic mental models, and fuzzy-trace theory) to explaina possible framing effect in capital budgeting decision contexts. AsArnold (1997, 62) has pointed out:
From a managerial accounting perspective, managers may re-view accounting information and make decisions that impact thefuture direction of the company. The initial interpretation of theinformation can determine the additional information that is con-sidered when making choices regarding the future….The impli-cations of framing to accounting environment is potentially quitesignificant.
In addition, systematic research is sparse on how managers use ac-counting information to ensure consistent business decisions (Brunsand McKinnon 1993, 85).3 Further, different theories represent differ-ent cognitive processes. An understanding of which theory can bestdescribe decision-making processes in accounting contexts would haveimplications for accountants in designing a better format for present-ing information to decision makers. For example, if prospect theorydominates, then one may consider providing multiple reference pointsin accounting reports or transforming a problem into a standard rep-resentation to prevent managers being affected by a framing effect(Arkes 1991; Hammond et al. 1998; Jamal et al. 1995). However, ifprobabilistic mental models dominate, other measures, such as pro-viding complete information rather than incremental information, maybe effective in assisting managers to make better decisions. If the fuzzy-trace theory can best describe the risky-choice framing effect in anaccounting context, numerical information should be presented care-fully so that decision makers cannot summarize it easily into vaguewordings. Finally, the distinction between framing effects and reflec-tion effects needs to be addressed so that accountants can focus onappropriate approaches for alleviating framing effects, not reflectioneffects, when designing accounting reports or information systems.This clarification may also shed light on previous mixed results, whichmay assist future accounting, auditing, and tax researchers in de-signing experiments to investigate framing effects.
3 The impact of incentives on performance has also been considered an importantfactor in decision making (Libby 1995; Libby and Luft 1993). However, Arkes (1991,495) suggested that framing effect is a “psychophysically based error.” Incentivesare effective in debiasing “strategy-based errors,” and ineffective in debiasing psy-chophysically based errors. In addition, Emby and Finley (1997) suggested that fram-ing effects seem to be data-related rather than effort-related and that the impact ofincentives on a data-related bias may be limited (Kennedy 1993, 1995). For thisreason, we decided to limit the scope of this study and not to include this factor inthe research design.
An Empirical Examination of Framing Effects 39
Two hundred seventy-one undergraduate business students par-ticipated in this study. Results suggest that the fuzzy-trace theorybetter explains potential framing effects on managerial decision mak-ing than does the often-applied prospect theory. Since the currentstudy is exploratory, further investigations are warranted.
The remainder of this paper is organized as follows. The next sec-tion briefly illustrates the framing effect and explains prospect theory.Two competing theories are introduced in the following section, whichleads to our research hypotheses. The fourth and fifth sections presentthe research method and the empirical results of Experiments 1 and2. Finally, discussions of the findings, implications and limitations ofthe study, and future research directions are highlighted in the lastsection.
PROSPECT THEORY AND FRAMING EFFECTSFraming Effect and the Classic Asian Disease Problem
The most well known example of a framing effect involves the clas-sic “Asian disease problem” (Tversky and Kahneman 1981, 453):
Problem 1:
Imagine that the U.S. is preparing for the outbreak of an unusualAsian disease, which is expected to kill 600 people. Two alterna-tive programs to combat the disease have been proposed. Assumethat the exact scientific estimates of the consequences of the pro-grams are as follows:
If Program A is adopted, 200 people will be saved.
If Program B is adopted, there is 1/3 probability that 600 peoplewill be saved and 2/3 probability that no people will be saved.
Which of the two programs would you favor?
Problem 2:
Imagine that the U.S. is preparing for the outbreak of an unusualAsian disease, which is expected to kill 600 people. Two alterna-tive programs to combat the disease have been proposed. Assumethat the exact scientific estimates of the consequences of the pro-grams are as follows:
If Program C is adopted, 400 people will die.
If Program D is adopted, there is 1/3 probability that nobody willdie and 2/3 probability that 600 people will die.
Which of the two programs would you favor?
For Problem 1 (using the positive wording “will be saved”), Tverskyand Kahneman (1981) reported that a clear majority of subjects (72percent) preferred saving 200 lives for sure. That is, in the domain of
40 Chang, Yen, and Duh
gains (will be saved), their subjects preferred the risk-averse ProgramA to the risky Program B that offers a 1/3 chance of saving 600 lives(28 percent). According to the expected utility theory (Friedman andSavage 1948), Programs C and D in problem 2 (using the negativewording “will die”) are identical to Programs A and B in problem 1.However, most subjects preferred Program D (78 percent) to ProgramC (22 percent). That is, framing the identical problem in different wayscan result in preference reversals. Tversky and Kahneman (1979) usedprospect theory as a framework to explain this phenomenon.
Prospect TheoryAccording to prospect theory, during a decision maker’s prelimi-
nary analysis of prospects, a psychological editing process (Phase I)takes place in order to organize the prospects, reformulate options,and simplify the subsequent evaluation and choice (Phase II). Duringthe editing phase, a perceived reference point is established to distin-guish gain from loss outcomes. The coding element of the editing pro-cess can be represented by a hypothetical value function. An S-shapedcurve that passes through a central reference point can be used todescribe prospect theory. This value function is predicted to be con-cave for gains and convex for losses, but it is also steeper for lossesthan for gains (Kahneman and Tversky 1979). The above conditionslead individuals to favor risk aversion for gains and risk seeking forlosses.
As mentioned earlier, outcomes are coded relative to a decisionmaker’s reference point. The reference point an individual adopts de-pends on how the problem is framed. For example, the positive prob-lem frame “will be saved” of Programs A and B in the classic Asiandisease Problem 1 would lead to an adjusted reference point that “thedisease will kill 600 people.” In this case, a decision maker perceivesthe outcomes of Programs A and B as gains (i.e., to save lives) and,having a tendency to be risk averse, would choose Program A (i.e.,200 people will be saved) because of the concavity of the S-shapedvalue function. In Problem 2, using the negative problem frame “willdie,” the reference point is not adjusted (i.e., nobody died so far), sothe decision maker perceives the outcomes of both Programs C and Das losses. If the decision maker is more risk seeking, due to the con-vexity of the S-shaped value function, he/she is predicted to chooseProgram D, because this alternative offers a one-third chance thatnobody will die.
The Distinction between the Framing Effect and the ReflectionEffect
Tversky and Kahneman (1981) suggested that pronounced prefer-ence reversals in risk result from the ways the choice/option out-comes are framed. Although this phenomenon is called a framing effect,
An Empirical Examination of Framing Effects 41
a reflection effect from prospect theory has been applied to explain it.Since Kahneman and Tversky (1979) did not differentiate reflectioneffects from framing effects, subsequent studies have used these twoterms interchangeably (e.g., O’Clock and Devine 1995; Schadewald1989), which may be the key reason for mixed results in research onframing effects (Fagley 1993; Kühberger 1995). This section describesboth effects and distinguishes one from the other.
In accordance with prospect theory, the S-shaped value functionis more relevant to problem domains (gain vs. loss) than to problemframes (positive vs. negative). The following two lottery experiments(Kahneman and Tversky 1979) demonstrate the impact of problemdomains.4
Problem 3a: Choose between (n = 95) (problem domain: gain)
A: Win $4,000 with probability .80, and $0 withprobability .20 [20%]
B: Win $3,000 [80%]
Problem 3b: Choose between (n = 95) (problem domain: loss)
A: Lose $4,000 with probability .80, and $0 withprobability .20 [92%]
B: Lose $3,000 [8%]
Kahneman and Tversky (1979) labeled the occurrence of reflection fromrisk-averse choices for a gain domain (Problem 3a) to risk-seekingchoices for a loss domain (Problem 3b) as a reflection effect. For areflection effect to occur, two gambles must be considered in whichone is generated from the other by replacing gains with losses, andvice versa. That is, a reflection effect does not involve the same gamble(Arkes 1991). The expected value of the problem in the gain domain ispositive, and that of the problem in the loss domain is negative. Whatappears to be a framing bias may in fact be a reflection effect if thedecision makers exhibit preference reversals in making choices fordifferent problems.
Compared with reflection effects, framing effects involve only oneproblem (e.g., Asian disease problem) with two frames (positive andnegative). For example, the actual problem domain of the Asian dis-ease problem is “loss.” It is the problem frames that illusively make aloss domain alternative appear to be in the gain domain under thepositive frame condition via the shift of reference point. As Li (1998)indicated, the term “framing effect” refers to changes in responsesfrom different descriptions of the same problem, whereas “reflectioneffect” refers to different responses because there are two problems.Table 1 analyzes the framing effect in the Asian disease problem.
4 The total number of respondents is denoted by n, and the percentage who choseeach option is indicated in the brackets.
42 Chang, Yen, and Duh
TA
BLE
1A
nal
ysi
s of
the
Fra
min
g E
ffec
t in
the
Asi
an D
isea
se P
roble
m
The
pro
ble
m:
Imagi
ne
that
the
U.S
. is
pre
pari
ng
for
the
outb
reak o
f an
un
usu
al A
sian
dis
ease
, w
hic
h is
expec
ted t
okill 600 p
eople
. Tw
o alt
ern
ati
ve p
rogr
am
s to
com
bat
the
dis
ease
have
bee
n p
ropos
ed.
Pro
ble
m D
om
ain
Act
ual
Pro
ble
mPro
gram
sPro
ble
m F
ram
eaPer
ceiv
edb
Dom
ain
c
A:
200 w
ill be
save
dPos
itiv
eG
ain
Los
s
B:1/3 c
han
ce t
hat
600 w
ill be
save
dan
d 2
/3 c
han
ce t
hat
0 w
ill be
save
dM
ixed
Gain
Los
s
C:400 w
ill die
Neg
ati
veLos
sLos
s
D:1
/3 c
han
ce t
hat
0 w
ill die
an
d 2
/3
chan
ce t
hat
600 w
ill die
Mix
edLos
sLos
s
Rev
ised
A:d
400 p
eople
will
not
be
save
d.
Neg
ati
veG
ain
Los
s
Rev
ised
C:d
200 p
eople
will
not
die
.Pos
itiv
eLos
sLos
s
aTh
e pro
ble
m f
ram
e (p
osit
ive
or n
egati
ve)
is d
ecid
ed b
y th
e ou
tcom
e of
th
e op
tion
an
d b
y th
e u
se o
f a n
egati
on “
not
.” T
hat
is,
the
pro
ble
m f
ram
e of
“200 (or
600) w
ill be
save
d”
or “
0 w
ill die
” is
pos
itiv
e, a
nd t
he
pro
ble
m f
ram
e of
“400 (or
600) w
ill die
” or
“0 w
ill be
save
d”
is n
egati
ve.
Wh
en u
sin
g th
e n
egati
on “
not
,” t
he
pro
ble
m f
ram
es o
f “2
00 w
ill n
ot d
ie”
an
d “
400 w
ill n
ot b
e sa
ved”
are
pos
itiv
ean
d n
egati
ve,
resp
ecti
vely
.b
Acc
ordin
g to
Kü
hber
ger
(1995), t
he
pro
ble
m d
omain
dec
isio
n m
aker
s per
ceiv
e (g
ain
or
loss
) dep
ends
only
upon
th
e w
ordin
g w
ith
rega
rd t
o th
e op
tion
ou
tcom
e (e
.g.,
“sa
ve”
or “
die
”). W
hen
usi
ng
the
wor
din
g “s
ave
d”
to d
escr
ibe
the
opti
on o
utc
ome,
dec
isio
n m
aker
sw
ill
per
ceiv
e th
e pro
ble
m d
omain
as
“gain
.” W
hen
usi
ng
the
wor
din
g “d
ie”
to d
escr
ibe
the
opti
on o
utc
ome,
dec
isio
n m
aker
s w
ill
per
ceiv
e th
e pro
ble
m d
omain
as
“los
s.”
cN
o m
att
er h
ow t
he
pro
ble
m is
fram
ed,
the
dom
ain
for
th
e A
sian
dis
ease
pro
ble
m s
hou
ld b
e co
nsi
der
ed a
los
s, s
ince
th
e ou
tbre
ak is
expec
ted t
o kill
600 p
eople
.d
Th
e re
vise
d v
ersi
on o
f Pro
gram
A (
Pro
gram
C)
use
s th
e n
egati
on “
not
” to
ch
an
ge t
he
pro
ble
m f
ram
e fr
om p
osit
ive
(neg
ati
ve)
ton
egati
ve (
pos
itiv
e).
An Empirical Examination of Framing Effects 43
Since reflection and framing effects are both predicted in prospecttheory by the S-shape of the value function, most studies view thesetwo terms similarly. However, these two effects are distinctly differ-ent. To be exact, a framing effect is the impact of how information ispresented to describe a specific problem, given the actual problemdomain is unchanged. Hence, framing effects manifest as a decisionbias (e.g., Emby and Finley 1997; Rutledge 1995), but a reflectioneffect requires different domains, irrespective of problem frames. Forexample, regardless of how the game is framed, the above Problems3a and 3b are distinctly different, with a gain domain and a loss do-main, respectively.
Since a framing effect is a perceptual phenomenon similar to anoptical illusion, whereas a reflection effect is not (Fagley 1993), dis-cerning the differences between these two effects is critical. When pro-viding financial information, accounting professionals should be alertto the format of the presentation to avoid causing possible framingeffects. Although a reflection effect does not involve a decision bias,accounting professionals should prepare remarks that will remind thedecision makers to view the alternatives as two distinctly separateproblems and to consider the different factors for each alternative.
In addition, as Arkes (1991) and Kühberger (1995) argued, framingand reflection effects have to be distinguished from each other so thatwe can properly interpret the results of research on framing effects.However, most of framing research in accounting thus far has not dis-tinguished framing effects from reflection effects. For example, inSchadewald’s (1989) study, subjects were not influenced by the fram-ing effect when the same taxation problem was used with different wordingto frame the options, but they did exhibit reflection effects using twopair of different decision problems. Schadewald (1989) did not separatethe two effects, which made it difficult to explain the mixed results.Likewise, the results of Sanders and Wyndelts (1989) partially sup-ported the reflection effect but did not support the framing effect.
Research Hypothesis Based on Prospect TheoryAccording to Kühberger (1995), the wording with regard to the
option outcome (e.g., “save” or “die”) decides the problem domain (i.e.,“gain” or “loss”) and relates to the reflection effect. On the other hand,the problem frame (i.e., “positive” or “negative”) depends on the use ofa negation “not,” which relates to the framing effect. That is, addingthe negation “not” to the sentence only changes the problem framefrom positive to negative, or vice versa, but it does not change theproblem domain (i.e., the gain or loss domain still holds). Using theAsian disease problem as an example, the possible combinations ofproblem domains and problem frames concerning the certain optionwith the negation “not” are illustrated in Table 1 as revised ProgramsA and C.
44 Chang, Yen, and Duh
In the classic Asian disease problem, it should be noted that thecombinations of the problem domain that decision makers perceivewith the problem frame are gain domain/positive frame and loss do-main/negative frame for Programs A and C, respectively. Therefore,the classic Asian disease test is ambiguous, because it confounds fram-ing effects and reflection effects. In order to separate the two effects, itis necessary to create another set of combinations (i.e., gain domain/negative frame and loss domain/positive frame). By comparing the re-sults of the two different combination sets, we will understand moreclearly the effects of framing and reflection.
According to prospect theory, decision makers will choose the cer-tain (risky) option over the risky (certain) option when the problem isperceived as a gain (loss) domain, regardless of the problem frame (posi-tive or negative). Accordingly, the hypotheses based on prospect theoryare generated as follows:
H1a: According to prospect theory, when the information ofa decision problem is stated in a gain domain/positiveframe, decision makers will choose the certain option(no risk) over the risky option. When the information ofa decision problem is stated in a loss domain/negativeframe, decision makers will choose the risky option overthe certain option.
H1b: According to prospect theory, when the information ofa decision problem is stated in a gain domain/negativeframe, decision makers will choose the certain option(no risk) over the risky option. When the information ofa decision problem is stated in a loss domain/positiveframe, decision makers will choose the risky option overthe certain option.
COMPETING THEORIESIn this section, we discuss two competing theories to explain fram-
ing effects: probabilistic mental models (Gigerenzer et al. 1991) andfuzzy-trace theory (Reyna and Brainerd 1991a, 1991b). Our discussionsof these two theories will lead us to develop the research hypotheses.
Probabilistic Mental ModelsGigerenzer et al. (1991) developed probabilistic mental models (PMM)
theory to explain and predict individuals’ behaviors related to overcon-fidence in judgments and decisions. Kühberger (1995) suggested thatPMM can also be used to explain the framing effect. According to PMM,individuals presented with a two-alternative task first attempt to con-struct a local mental model (LMM) of the task, then use it to solve the
An Empirical Examination of Framing Effects 45
problem using long-term memory and elementary logical operations.In general, a LMM can be successfully constructed if (1) precise fig-ures can be retrieved from long-term memory to compare alternatives,(2) ranges/features of the information regarding the alternatives donot overlap, or (3) elementary logical operations, such as exclusion,can compensate for missing knowledge. If the problem cannot be solveddirectly using a LMM, then a PMM is constructed using probabilisticinformation generated from long-term memory. Most accounting andmanagement decisions would require the use of a PMM, since the sec-ond and third requirements for the use of a LMM usually are not metin solving business problems. The PMM theory suggests that a deci-sion maker solves a problem by applying inductive inference, i.e., byputting the specific decision task into a larger context (Gigerenzer etal. 1991, 507).
According to the theory of PMM, a reference class of the decisionproblem and a network of decision variables, in addition to the targetvariable, are used to perform an indirect, frequency-based inference.In other words, to make a decision, an individual first constructs areference class for the specific problem. Recall the four programs inthe Asian disease problem. For this particular problem, the referenceclass could be “programs for fighting disasters” (Kühberger 1995).Besides the target variable (i.e., to save lives), important variables inthe problem may include time, newly developed knowledge, skills and/or medicines, additional resources, etc. Since Program A indicates that200 people will be saved but does not specify the possible outcome forthe remaining 400 people, Program A leaves room for individuals tobuild their own PMM.
After the subjects have read the problem and compared ProgramsA and B regarding how many lives will be saved, they may concludethat, as time progresses, new cures for the disease could be identifiedand additional resources could be allocated to deal with the disaster.Therefore, after an initial number of people have been saved, someadditional people may be saved as well. That is, under the wording“will be saved,” Programs A and B may be stated as follows:
Program A: 200 people will be saved and some more may be savedlater.
Program B: There is 1/3 probability that 600 people will be savedand 2/3 probability that no people will be saved.
The problem frame of Program A is positive, while that of Program B ismixed, which may be the reason why subjects under this wordingsituation prefer Program A to Program B. Program A allows for a pos-sibility that additional people may be saved. On the other hand, whenthe subjects compare Programs C and D (under “will die” wording),they perceive the fact that under Program C 400 people will die for
46 Chang, Yen, and Duh
sure. In addition, the probability could be low that new knowledge willbe acquired to save people as time progresses. Therefore, additionalpeople may die after a relatively large portion of people die. Therefore,Programs C and D may be stated as follows:
Program C: 400 people will die and some more may die later.
Program D: There is 1/3 probability that nobody will die and 2/3probability that 600 people will die.
This could explain why subjects prefer Program D (mixed problemframe) to Program C (negative problem frame). Thus, due to the ex-perimental design (gain domain/positive frame for Program A, andloss domain/negative frame for Program C) of the classic Asian dis-ease problem, either the reflection effect or PMM effect could accountfor the framing effect. However, it should be noted that a necessarycondition of PMM in explaining the framing effect is that the certainoption is described with incomplete information, which leaves roomfor decision makers to consider other possible variables relevant tothe problem. In addition, the certain option is described using differ-ent problem frames (positive or negative), which may lead decisionmakers to opposite views of the same problem. Opposite to prospecttheory, the theory of PMM focuses on the problem frame, and the prob-lem domain (gain or loss) is irrelevant. Therefore, in order to test thedescriptive abilities of prospect theory and PMM, another set of com-binations of problem domains and problem frames is designed (i.e.,gain domain/negative frame vs. loss domain/positive frame). Accord-ingly, the hypotheses based on the PMM theory are as follow:
H2a: According to the PMM theory, when the information ofa decision problem is stated in a gain domain/positiveframe, decision makers will choose the certain option(no risk) over the risky option. When the information ofa decision problem is stated in a loss domain/negativeframe, decision makers will choose the risky option overthe certain option.
H2b: According to the PMM theory, when the information ofa decision problem is stated in a gain domain/negativeframe, decision makers will choose the risky option overthe certain option (no risk). When the information of adecision problem is stated in a loss domain/positiveframe, decision makers will choose the certain optionover the risky option.
Fuzzy-Trace TheoryBased on previous research on the relationship between memory
and reasoning, Reyna and Brainerd (1990) derived fuzzy-trace theory
An Empirical Examination of Framing Effects 47
(FFT). This theory differs from prospect theory, which searches for apsychophysical function for probabilities and outcome values, in thatFTT assumes individuals prefer to reason using simplified representa-tions of information (i.e., gist), as opposed to exact details (Reyna andBrainerd 1991a). Thus, according to FTT, while individuals are encod-ing verbatim information, they extract global patterns from the infor-mation presented and then mentally represent the decision problemat different levels of specificity. This fuzzy-to-verbatim continuum ofrepresentations allows decision makers the latitude to incorporate theirown personal, fuzzily processed preferences into the options. Typi-cally, an overall impression of the task is derived to determine whethergist-based processing is feasible (Ashcraft and Battaglia 1978; Gelman1972). For example, would estimation, as opposed to calculation, suf-fice to solve a numerical problem? That is, multiple representationsavailable in a task can be ordered according to a hierarchy of gist, andgist-based reasoning is predicted to operate at the vaguest level in thehierarchy that permits discrimination, given the constraints of thetask.
Reyna and Brainerd (1991a, 1995) used FTT to explain the classicframing effect. When quantitative information is available, “relationalgist” (i.e., the notion that one option has “more” or “less” than theother) is automatically extracted. That is, individuals generate a no-tion based on “more” or “less” vs. “some” to distinguish between theoptions. When the options involve null outcomes (e.g., nobody saved),however, the comparison becomes “some” vs. “none” or “presence” vs.“absence” of an attribute. Note that, as with numerical outcomes, prob-abilities are represented dichotomously—certainty, a single possibleoutcome, as opposed to uncertainty, more than one possible outcome.This is presumed to occur even if people have access to more concreteinformation. Hence, according to the FTT, the vague distinctions ofthe options in the Asian disease problem can be stated as follows:
Program A: Some people will be saved.Program B: Some people will be saved or no one will be saved.
Program C: Some people will die.
Program D: Nobody will die or some people will die.
Based on FTT, to make a choice between Programs A and B, “somepeople will be saved” is common to both alternatives, and the differencecenters on “no one will be saved.” Hence, individuals prefer Program A.In comparing Program C to Program D, “some people will die” is com-mon to both alternatives, therefore, individuals focus on the differentpart, that “nobody will die,” and prefer Program D. As is evident in thestudies of Reyna and Brainerd (1991a, 1995), removing all of the num-bers from the classical disease problems and replacing them with vaguephrases does not eliminate the framing effect. Indeed, under these
48 Chang, Yen, and Duh
circumstances, the framing effects were more significant in magni-tude than when numbers were used. That is, the reflected framingdata are explained as being the result of qualitative distinctions, not ofthe detailed information or of calculations.
Stone et al. (1994) concluded that fuzzy-trace theory better ex-plains framing effects. They claimed and confirmed that people de-scribe the difference between risks of .006 and .003 as “low, butsignificant, risk.” On the other hand, people perceive the differencebetween risks of .000006 and .000003 as “essentially nil risk.” Thatis, a fuzzy-processing preference serves to drive the representation ofquantities down to a hierarchy of gist. In addition, contemporary re-searchers in the area of judgment and decision making have used FTTto explore reasoning involved with conditional probability (Wolfe 1995),social stereotypes (Davidson 1995), motivational biases (Klaczynskiand Fauth 1997), and various types of deduction, including causalreasoning (Klaczynski and Narasimham 1998).
Note that when the negation “not” is added to the certain option(e.g., the revised Programs A and C of Table 1), the certain option andthe risky option have no identical part; thus, individuals cannot sim-plify the decision options to gist levels. Under this situation, FTT indi-cates that decision makers must exert additional cognitive effort toreach a decision. Klaczynski and Fauth (1997) adopted FTT to exploreself-serving bias. They found that when evidence contradicts cher-ished beliefs about self, subjects struggle to overcome gist-based in-terference in order to engage in quantitative reasoning. Consequently,the same subjects exhibit more advanced reasoning (e.g., applyingstatistical rules) in decision making. For more complex decision prob-lems in accounting or management, one can assume that decisionmakers may have to exercise reasoning at the numerical level whenoptions are different and cannot be simplified to gist levels.
According to FTT, if additional effort is exerted to make a decisionand the options suggest the same expected value at a numerical level,then individual differences in risk preference may moderate the impactof framing effects. Wang (1996) indicates that when a decision maker’srisk preference is weak, she/he may become more sensitive to the fram-ing effect; however, when a decision maker’s risk preference is strong,she/he is more immune to the framing manipulation. Of interest in amanagerial accounting context is the question of the relationship be-tween framing effects and individuals’ risk preferences when the al-ternatives in a decision problem yield the same expected value.
In summary, fuzzy-trace theory predicts that the qualitativerelationships among numerical values, rather than the values them-selves, govern the decision unless the individual cannot simplify thedecision options due to complex information. In other words, if cer-tain options are described using gain domain/positive frame or loss
An Empirical Examination of Framing Effects 49
domain/negative frame (e.g., Programs A and C of Table 1), FTT predictsthat individuals will make decisions at the gist level, thus the framingeffect will exist. Therefore, the classic Asian disease test is ambiguousbecause it confounds not only the effects of reflection and PMM, but offuzzy-trace, as well.
According to FTT, using the negation “not” in the certain options(e.g., the revised Programs A and C of Table 1) provides informationthat individuals cannot simplify to gist levels but which requires addi-tional cognitive effort to reach a decision. In this case, the framingeffect will be absent. Accordingly, our hypotheses based on the fuzzy-trace theory are as follows:
H3a: According to FTT, when the information of a decisionproblem is stated in a gain domain/positive frame, de-cision makers will choose the certain option (no risk)over the risky option. When the information of a deci-sion problem is stated in a loss domain/negative frame,decision makers will choose the risky option over thecertain option.
H3b: According to FTT, when the information of a certainoption is stated in a gain domain/negative frame or ina loss domain/positive frame, the framing effect will beabsent.
RESEARCH METHODExperiment 1 is designed to show the robustness of the classic
framing effect in a managerial accounting context and to see whetherthe three theories have equal explanatory ability to predict this phe-nomenon. The second experiment investigates which theory best de-scribes decision behavior in an accounting context.
As Sanders and Wyndelts (1989) indicated, since the manipulationof the independent variables in framing studies generally involved merechanges in the wording of the scenarios, a within-subjects design wasnot appropriate. We employed a 2 × 3 between-subjects design acrosstwo experiments to avoid a possible demand effect of using a within-subjects design (Pany and Reckers 1987). Two variables were manipu-lated: problem domain (gain vs. loss) and problem frame (positive,negative, and mixed). Table 2 summarizes the research design andthe experimental material and compares them with those of the clas-sic Asian disease problem. The four situations included in Experi-ment 1 are gain domain/positive frame, gain domain/mixed frame,loss domain/negative frame, and loss domain/mixed frame. Experi-ment 2 presents the business situations as gain domain/negativeframe, gain domain/mixed frame, loss domain/positive frame, andloss domain/mixed frame.
50 Chang, Yen, and DuhT
AB
LE
2A
Sum
mar
y o
f E
xper
imen
tal
Des
ign a
nd M
ater
ial
Com
par
ed w
ith t
he
Cla
ssic
Asi
an D
isea
se P
roble
m
Per
ceiv
edSta
tem
ents
Th
e C
lass
ic A
sian
Dis
ease
Pro
ble
mPro
ble
m(A
lter
nat
ives
)Pro
ble
mE
xper
imen
tal
Mat
eria
lD
om
ain
Fra
me
Exper
imen
t
Th
e U
.S.
is p
repari
ng
for
the
Th
e co
mpan
y’s
curr
ent
pol
luti
onou
tbre
ak o
f an
un
usu
al A
sian
con
trol
sys
tem
no
lon
ger
mee
tsdis
ease
, w
hic
h is
expec
ted t
oth
e m
inim
um
req
uir
emen
t. T
he
kill 600 p
eople
. C
hoi
ce o
f on
eco
mpan
y m
ay b
e su
bje
ct t
o a
from
tw
o op
tion
s.$300,0
00 p
un
itiv
e fin
e. C
hoi
ceof
on
e fr
om t
wo
opti
ons.
1C
erta
in200 p
eople
will be
save
d.
Th
e co
mpan
y w
ill, f
or c
erta
in, b
e ab
leG
ain
Pos
itiv
e1
to s
ave
$100,0
00 in
pu
nit
ive
fin
e.
2R
isky
Th
ere
is 1
/3 p
robabilit
y th
at
Th
ere
is 1
/3 p
robabilit
y th
at
the
Gain
Mix
ed1 a
nd 2
600 p
eople
will be
save
d a
nd
$300,0
00 p
un
itiv
e fin
e w
ill be
2/3 p
robabilit
y th
at
no
peo
ple
save
d a
nd 2
/3 p
robabilit
y th
at
will be
save
d.
$0 p
un
itiv
e fin
e w
ill be
save
d.
3C
erta
in400 p
eople
will die
.Th
e co
mpan
y w
ill, f
or c
erta
in,
be
Los
sN
egati
ve1
subje
ct t
o a
$200,0
00 p
un
itiv
e fin
e.
4R
isky
Th
ere
is 1
/3 p
robabilit
y th
at
Th
ere
is 1
/3 p
robabilit
y th
at
the
Los
sM
ixed
1 a
nd 2
nob
ody
will die
an
d 2
/3
com
pan
y w
ill be
subje
ct t
o $0
pro
babilit
y th
at
600 p
eople
pu
nit
ive
fin
e an
d 2
/3 p
robabilit
yw
ill die
.th
at
the
com
pan
y w
ill be
subje
ctto
a $
300,0
00 p
un
itiv
e fin
e.
5C
erta
in400 p
eople
will
not
be
save
d.
Th
e co
mpan
y w
ill, f
or c
erta
in,
Gain
Neg
ati
ve2
not
be
able
to
save
$200,0
00
in p
un
itiv
e fin
e.
6C
erta
in200 p
eople
will
not
die
.Th
e co
mpan
y w
ill, f
or c
erta
in,
Los
sPos
itiv
e2
not
be
subje
ct t
o a $
100,0
00
pu
nit
ive
fin
e.
An Empirical Examination of Framing Effects 51
MaterialsThe case used in both experiments is a capital investment sce-
nario. The subjects were asked to assume the role of controller at ahypothetical company (TilTec Inc.). The experiments asked subjects tochoose between two options (A, a certain option, and B, a risky option)for purchasing new equipment to meet newly announced environmentalprotection standards.5 The subjects could write comments in spacesprovided in the case.
Experiment 1In Experiment 1, the case scenario states that, due to recently
raised environmental protection standards, the company’s current pol-lution control system no longer meets the minimum requirement. If noimprovement is made soon, the company may be subject to a $300,000punitive fine. The controller must choose between two options, A andB. The equipment in each option has the same useful life. Under bothoptions A and B, costs related to purchases, operations, and othermiscellaneous expenses, such as maintenance, are also identical.
For the gain domain condition, option A (positive frame) is statedas “If the equipment under option A is purchased, TilTec will, for cer-tain, be able to save $100,000 out of the $300,000 total punitive fine,”and option B (mixed frame) is stated as “If the equipment under op-tion B is purchased, there is a one-third probability that TilTec willsave all of the $300,000 punitive fine and a two-thirds probabilitythat TilTec will save $0 of the punitive fine.” For the loss domain con-dition, option A (negative frame) is stated as “If the equipment underoption A is purchased, TilTec will, for certain, be subject to a $200,000punitive fine,” and option B (mixed frame) is stated as “If the equip-ment under option B is purchased, there is a two-thirds probabilitythat TilTec will be subject to a $300,000 punitive fine and a one-thirdprobability that TilTec will be subject to $0 punitive fine.” That is, theexpected values of both options are the same. For both domains, choos-ing option A suggests a risk-averse attitude, while choosing option Bis risk seeking.
To test the predictive ability of each theory concerning the classicframing effect in an accounting-related decision, we made the abovecapital investment problem correspond with the Asian disease prob-lem. Table 2 provides the comparison between the experimentalmaterial and the classic Asian disease problem. In Table 2, statements1, 2, 3, and 4 are formally and structurally equivalent to the Asiandisease problem and were used in Experiment 1. The first two state-ments are for the gain domain, and statements 3 and 4 are for the
5 After performing the task, they were also asked to express their level of confidence intheir decisions (on a seven-point Likert scale, with 1 as “not confident at all” and 7as “totally confident”).
52 Chang, Yen, and Duh
loss domain. The certain options, statements 1 and 3, are in the gaindomain/positive frame and the loss domain/negative frame, respec-tively (see Table 2). Since the gain domain confounds with the positiveframe and the loss domain confounds with the negative frame, thisdesign does not separate the framing effect from the reflection effect.Hence, based on our discussion in the previous theory sections, pros-pect theory, PMM theory, and FTT all predict the presence of the clas-sic framing effect in Experiment 1 (see Table 3).
Experiment 2Experiment 2 is created to differentiate the explanatory power of
the three theories. The case scenario used in this experiment is thesame as Experiment 1, with some specific revisions of the certain op-tions. Recall that, in predicting framing effects, prospect theory is con-cerned with problem domains (gain vs. loss), while the theory of PMMfocuses on problem frames (positive vs. negative) and whether com-plete information is provided in the decision problem or there is roomfor individuals to take other relevant variables into consideration. TheFTT focuses on whether the stated options can be simplified or thedecision problem must be solved at a more complex level, such as bythe use of numerical computations. In Experiment 2, we design theoption statements in a manner that allows us to separate the effect ofthe problem domain from the problem frame and in such a way thatthe information cannot be simplified. This allows us to see the dis-tinct prediction of each theory. The empirical results will provide usevidence as to which theory can best describe the subjects’ behaviors.
In this experiment, we maintain the same statements for the riskyoptions (statements 2 and 4 in Table 2) and revise those for the cer-tain options. Statement 5 is a revised version of statement 1 withoutchanging the fact in the statement (see Table 2). The problem domainfor both statements is the same (gain domain), but the problem frameis changed from positive to negative. Similarly, statement 6 is modi-fied from statement 3 to change the problem frame from negative topositive with a loss problem domain. The purpose of these changes isto make the revised statements for the certain options (statements 5and 6) in the gain domain/negative frame and the loss domain/posi-tive frame. Since the revised statements no longer confound the gaindomain with the positive frame and the loss domain with the negativeframe, we are able to create different predictions under each theory.
According to prospect theory, which is based on the reflection ef-fect (problem domain), individuals are predicted to prefer the certainoption (statement 5) over the risky option (statement 2) in a gain do-main and to prefer the risky option (statement 4) over the certain op-tion (statement 6) in a loss domain.
Concerning the PMM theory, statement 2 provides complete infor-mation, but statement 5 does not. Statement 5 reads, “The company
An Empirical Examination of Framing Effects 53
TA
BLE
3H
ypoth
eses
and R
esult
s
Exper
imen
tH
ypoth
esis
an
d P
redic
tion
Support
ing
Th
eory
Con
firm
ed o
r N
ot?
H1a:
Nu
mber
of
subje
cts
choo
sin
g A
lter
nati
ve 1
Pro
spec
t th
eory
Yes
> N
um
ber
of
subje
cts
choo
sin
g A
lter
nati
ve 2
;N
um
ber
of
subje
cts
choo
sin
g A
lter
nati
ve 4
> N
um
ber
of
subje
cts
choo
sin
g A
lter
nati
ve 3
H2a:
Nu
mber
of
subje
cts
choo
sin
g A
lter
nati
ve 1
PM
MY
es1
> N
um
ber
of
subje
cts
choo
sin
g A
lter
nati
ve 2
;N
um
ber
of
subje
cts
choo
sin
g A
lter
nati
ve 4
> N
um
ber
of
subje
cts
choo
sin
g A
lter
nati
ve 3
H3a:
Nu
mber
of
subje
cts
choo
sin
g A
lter
nati
ve 1
FTT
Yes
> N
um
ber
of
subje
cts
choo
sin
g A
lter
nati
ve 2
;N
um
ber
of
subje
cts
choo
sin
g A
lter
nati
ve 4
> N
um
ber
of
subje
cts
choo
sin
g A
lter
nati
ve 3
H1b:
Nu
mber
of
subje
cts
choo
sin
g A
lter
nati
ve 5
Pro
spec
t th
eory
No
> N
um
ber
of
subje
cts
choo
sin
g A
lter
nati
ve 2
;N
um
ber
of
subje
cts
choo
sin
g A
lter
nati
ve 4
> N
um
ber
of
subje
cts
choo
sin
g A
lter
nati
ve 6
H2b:
Nu
mber
of
subje
cts
choo
sin
g A
lter
nati
ve 5
PM
MN
o2
< N
um
ber
of
subje
cts
choo
sin
g A
lter
nati
ve 2
;N
um
ber
of
subje
cts
choo
sin
g A
lter
nati
ve 4
< N
um
ber
of
subje
cts
choo
sin
g A
lter
nati
ve 6
H3b:
Nu
mber
of
subje
cts
choo
sin
g A
lter
nati
ve 5
FTT
Yes
= N
um
ber
of
subje
cts
choo
sin
g A
lter
nati
ve 2
;N
um
ber
of
subje
cts
choo
sin
g A
lter
nati
ve 4
= N
um
ber
of
subje
cts
choo
sin
g A
lter
nati
ve 6
54 Chang, Yen, and Duh
will, for certain, not be able to save $200,000 punitive fine.” Accordingto the PMM theory, decision makers consider more variables than justthe target variable when information is not complete. For example,they may suspect that the government will eventually come down hardon companies that do not significantly improve pollution controls.Therefore, subjects are predicted to interpret statement 5 as “The com-pany will, for certain, not be able to save $200,000 punitive fine now,and it may not be able to save the rest of the fine.” That is, accordingto PMM, individuals are predicted to prefer statement 2 over state-ment 5. On the other hand, statement 6 reads, “The company will, forcertain, not be subject to a $100,000 punitive fine.” This could beinterpreted as, “The company will, for certain, not be subject to a$100,000 punitive fine, and it will not be subject to the rest of thefine.” Hence, individuals are predicted to prefer statement 6 over state-ment 4.
The FTT expects individuals to simplify the decision problem togist level. When the problem cannot be simplified, the FTT indicatesthat individuals will solve the problem at a more complex level usingnumerical computations. If the numerical results of the two alterna-tives are the same, then risk preference is predicted to influence thefinal decision. In Experiment 2, since the two options have no similarpart, they cannot be simplified to make an easy decision. That is, tomake a comparison, individuals must solve the problem at the nu-merical level by calculating the expected value for each option. Sincethe expected value is the same for both options, the driving factor ispredicted to be the individual’s risk attitude. Table 3 summarizes thehypotheses and predictions of each theory.
SubjectsEighty-six students6 participated in Experiment 1, and 185 students7
participated in Experiment 2. All subjects were recruited from two met-ropolitan, state universities on the West Coast and had completed amanagerial accounting course. During the recruitment, subjects weretold that they would be engaging in a judgmental task related to a busi-ness decision. The subjects were also assured that their responses would
6 The subjects had an average of 1.48 years of accounting-related work experience,with a standard deviation of 2.44 years. The average age for this group was 26.48 years.On a seven-point Likert scale, the subjects were asked to evaluate the experimentalmaterials. The statistics indicate that the case was interesting (with mean and stan-dard deviation 4.70 and 1.15, respectively) and understandable (with mean andstandard deviation 4.72 and 1.21, respectively).
7 The subjects had an average of 0.22 years of accounting-related work experience,with a standard deviation of 0.84 years. The average age of subjects in this group was24.82 years. The subjects indicated that the case was interesting (with mean andstandard deviation 4.89 and 1.19, respectively) and understandable (with meanand standard deviation 5.00 and 1.23, respectively).
An Empirical Examination of Framing Effects 55
be handled confidentially and that data would be analyzed only ingroups. The subjects were not told the nature or the objective of theexperiments.
Experimental ProceduresUpon arrival at the experiment sites, subjects were randomly as-
signed to each treatment and given a one-page instruction sheet. Thissheet provided a short description of the experiment and stated thatthere are no right or wrong answers to the experimental task. Thesubjects were instructed to make their own judgments based on thecase scenario. Then, the subjects were asked to study the managerialcase and make a choice between two alternatives. Finally, the sub-jects completed a post-experiment questionnaire that contained bothdemographic and risk-attitude questions. Subjects took, on average,approximately 20 minutes to complete the experiment.
RESULTSExperiment 1
The purpose of Experiment 1 was to test the robustness of theclassic framing effect in a managerial accounting context. As Table 4indicates, when the options were stated in the gain domain, 29 and13 students chose options A (the certain option) and B (the risky op-tion), respectively. That is, approximately 31 percent of the subjectspreferred a risky alternative. On the other hand, when the optionswere stated in the loss domain, 16 and 27 students chose options Aand B, respectively. That is, approximately 63 percent of subjectstended to be risk seeking in that situation. As shown in Table 4, theChi-square result indicates a significant difference (Chi-square = 8.645,p < 0.003).8 This finding supports H1a, H2a, and H3a that prospecttheory, PMM, and FTT all predict the classic framing effect in a mana-gerial accounting context.
8 Since expressing “no difference” between alternatives is not permitted in our experi-ments, subjects could have been forced to choose an alternative. We next designed atest to confirm our Chi-square result. The dependent variable of this test is createdby multiplying the subject’s confidence level (1 to 7) with the option chosen. The norisk/certain option is coded 1, and the risky option is coded –1. If the subjects wereindeed coerced into making their choices, their confidence levels about their deci-sions should be low (1 or 2 on a seven-point Likert scale). Hence, no matter whichoption is chosen, according to our design, the dependent variable from coerced sub-jects should be low. Thus, the result of this ANOVA test may be insignificant, eventhough the Chi-squared is significant. On the other hand, if the subjects were notforced to make their choices, then their confidence levels should not be low. There-fore, the results of the ANOVA should be the same as the result of the Chi-square.Such is the case in experiment 1 (F = 11.693, p < 0.001). That is, the ANOVA resultconfirms our Chi-square findings in Experiment 1.
56 Chang, Yen, and Duh
TA
BLE
4N
um
ber
(Per
centa
ge)
of
Subje
cts
Choosi
ng
Eac
h A
lter
nat
ive
in E
xper
imen
ts 1
and 2
Fra
min
gSta
tist
ical
Tes
ts
Posi
tive
Neg
ativ
eC
ross
Tab
ula
tion
sF
ram
ing
Cer
tain
Opti
on
Ris
ky O
pti
on
Cer
tain
Opti
on
Ris
ky O
pti
on
Ch
i-sq
uar
ep
Eff
ect
Exp
erim
ent
129 (
69.0
%)
13 (
31.0
%)
16 (
37.2
%)
27 (
62.8
%)
8.6
45
0.0
03
Yes
Exp
erim
ent
247 (
51.1
%)
45 (
48.9
%)
51 (
54.8
%)
42 (
45.2
%)
0.2
61
0.6
60
No
An Empirical Examination of Framing Effects 57
Experiment 2Experiment 2 is designed using a revised version of the classic Asian
disease problem to differentiate among the descriptive abilities of pros-pect theory, PMM, and FTT with regard to the impact of informationpresentation on decision makers’ behaviors. As reported in Table 4,when the options in Experiment 2 were stated in the gain domain, 47students chose the certain option, A, and 45 chose the risky option, B.That is, in such a situation, about 51 percent of the subjects were riskaverse, and 49 percent were risk seeking. On the other hand, when theinformation was stated in the loss domain, 51 subjects chose option Aand 42 chose option B. That is, a total of 55 percent of the subjectswere risk averse, and 45 percent were risk seeking. The Chi-squareresult indicates no significant difference (Chi-square = 0.261, p < 0.660).9
As shown in Tables 3 and 4, this result supports the predictions of FTT.Our finding is consistent with that of Stone et al. (1994), which sug-gests that individuals reason on the basis of simplified representationsrather than on the literal information available.
Additional Analysis (Framing Effects and Risk Preference)Fuzzy-trace theory indicates that if different options cannot be sim-
plified to gist-vague levels, then decision makers may need to exercisereasoning at the numerical level. If the options’ outcomes result in thesame numerical level, then the decision is predicted to be affected bythe decision makers’ risk propensity. Wang (1996) and Zickar andHighhouse (1998) found that individuals’ risk preferences, as well asframing, affected their subjects’ decision making. To confirm this pre-diction of FTT, we further analyzed subjects’ choices in both Experi-ments 1 and 2.
Using subjects’ choices as the dependent variable, we conductedANCOVA tests with the subjects’ risk preferences10 as the covariate.The results shown in Table 5 support the prediction that if an option isstated in such a way that it can be simplified (i.e., Experiment 1), thenthe framing effect is present, and the individual decision maker’s riskpreference is not the driving factor in making the decision (Panel A ofTable 5, p < 0.779). On the other hand, when the option cannot be
9 According to Kirk (1995), the size of the effect (f = 0.355) in Experiment 1 is betweenmedium (f = 0.25) and large (f = 0.40). If we set (1) the level of significance α = 0.05,(2) power 1 – β = 0.9, and (3) medium effect size (f = 0.25), the necessary sample sizeof Experiment 2 should be 168 (for details, please refer to Kirk [1995, 187 and TableE.13]). Since the actual sample size of Experiment 2 is 185, our finding is not simplya consequence of low statistical power. The power of Experiment 2 is higher than0.9.
10 We used both self-reported risk attitudes and a validated risk propensity test (Koganand Wallach 1964) to measure subjects’ risk preferences. Pearson correlation sug-gested these two measures are correlated (r = .261, p < 0.001). In ANCOVA tests, weused the measurement resulting from the risk propensity test.
58 Chang, Yen, and Duh
simplified and the numerical conclusions reveal an equal outcome ofthe options (i.e., Experiment 2), the framing effect is absent and theindividual’s risk preference drives the decision (Panel B of Table 5,p < 0.003).
DISCUSSIONS AND CONCLUSIONSTo determine which theory can best explain the well-known fram-
ing effect in making an accounting-related decision, this study exam-ined three competing theories, prospect theory, probabilistic mentalmodels, and fuzzy-trace theory. The results suggest that FTT best de-picts the phenomenon of framing effects on decision makers’ behaviorin an accounting context, though prospect theory has been appliedmost commonly.
Consistent with Simon’s (1956) suggestion that individuals areusually bound by human cognitive limitations, FTT suggests that de-cision makers prefer to reason using greatly simplified, summarizedrepresentations of information, as opposed to exact details (Reyna andBrainerd 1991a). That is, individuals tend to process information us-ing qualitative patterns rather than precise quantities such as prob-ability values or numerical outcomes. FTT further suggests thatdecision makers tend to process information at a numerical level ifthe information cannot be pictured qualitatively (i.e., at a gist level).When the numerical outcomes of the different alternatives are identi-cal, the driving factor in making the decision is an individual’s riskpreference. The results from this study support the predictions basedon FTT. We observed framing effects in a managerial accounting con-text when the information presented could be extracted to a gist level.On the other hand, framing effects were absent when the informationcould not be simplified. The subjects’ written comments also support
TABLE 5ANCOVA Results on Subjects’ Choices
with Risk Preference as Covariate
Panel A: Experiment 1
Source Mean Square df F p
Risk Preference 2.041 1 0.079 0.779Framing 298.635 1 11.591 0.001Residual 25.765 82
Panel B: Experiment 2
Source Mean Square df F p
Risk Preference 230.771 1 8.860 0.003Framing 0.991 1 0.038 0.846Residual 26.039 181
An Empirical Examination of Framing Effects 59
the FTT. For example, in Experiment 2, one participant who choseoption A (the certain option) stated that option B has the same out-come as option A in terms of the expected amount of punitive fine tobe paid, but it is far more risky than option A. That is, she/he madethe choice based on a personal risk preference. Another comment reads,“At first glance, both options are the same, but the second is risky.”
There are several implications from this study. The results of Ex-periment 2 support fuzzy-trace theory by showing that if individualscannot simplify the information, then they process the information atthe numerical level. One implication of this finding is that it might bepossible to force decision makers to process information at a higherlevel by presenting information in certain formats (cf., Libby and Luft1993). Since decision biases may depend on how information has beenprepared by accountants, a reporting system should be designed toalleviate framing effects by presenting financial information that can-not be easily summarized into vague wordings.
More importantly, when auditors exercise their professional judg-ments, they are also influenced by how information (e.g., financialinformation or audit evidence) has been presented to them. Becausemanagers are allowed to use their own discretion in framing state-ments for annual reports that communicate company information tousers of financial statements (Healey and Palepu 1993), it is possiblefor managers to mislead users of financial statements or to mask im-proper actions such as fraud (Johnson et al. 1991). An understandingof framing effects may assist auditors in detecting questionable repre-sentations (Jamal et al. 1995).
Finally, the finding of a relationship between framing effects andindividuals’ risk preferences calls attention to the dynamic features offraming effects. As Wang (1996) suggested, the possible interactioneffect between risk preference and frames may be contingent uponthe specific task domain. Our results echo Arnold’s (1997) speculationthat the framing effects found in previous auditing research (e.g.,McMillan and White 1993) may be related to the various levels of riskinherent in the audit environment. Future studies investigating theantecedent conditions that determine the direction and presence of fram-ing effects could provide useful insights to the accounting profession.
There are some limitations of this study. First, as with other stud-ies in this line of research, the external validity might be restricted.Since subjects in this study were undergraduate business majors, theirexperience in making capital investment decisions is limited. To im-prove external validity, future studies could recruit subjects with task-specific experience. In addition, since the scenario of our study wassimplified, researchers should also consider developing cases with morerealistic/complex scenarios to investigate whether the fuzzy-tracetheory is better than other competing theories for explaining framingeffects.
60 Chang, Yen, and Duh
Third, the wording used in our experiments to manipulate differ-ent decision domains may not have been salient to the subjects. Forexample, in our second experiment, it is possible that subjects wouldhave been more sensitive to wording “penalized with” rather than “sub-ject to…punitive fine.” For example, Luft (1994) found that employeesare more likely to accept incentive contracts described in bonus termsrather than contracts that appear identical except for being describedin penalty terms. She explained this phenomenon in terms of the hu-man information-processing costs of communicating and evaluatingthe contract terms. Therefore, our results should be interpreted withcaution.
In addition, we did not directly ask the subjects to indicate theirperceived problem domain vs. problem frame during the experiments.Instead, we provided them with spaces for overall comments on thecase scenario. There were 157 subjects (57.93 percent) who commentedon the case scenario (44 out of 86 for Experiment 1 and 113 out of 185for Experiment 2). In Experiment 1, 8 out of 44 subjects (18.18 percent)mentioned that their calculations indicated two options having the sameexpected value. In Experiment 2, 39 out of 113 subjects (34.51 percent)indicated that the expected values of options A and B are the same. Thedifference is significant (t = 2.007, p < 0.047), which indirectly supportsFTT’s prediction that individuals tend to solve a problem at the numeri-cal level if they cannot simplify the problem. Concerning PMM’s predic-tion on information completion, two (three) subjects in Experiment 1(2) stated the descriptions of the option outcomes as not enough or notcomplete. This result does not seem strong enough to support the pre-dictions of PMM. It is critical that future studies conduct manipulationchecks to directly confirm the predictions of these theories.
Future research should clearly distinguish between the variousaspects of framing, such as standard risky choice, attribute framing,and goal framing, in experimental designs. In addition, studies in at-tribute framing that investigate decision types, rather than simplymaking choices between options, may warrant further examination.For instance, consider auditing tasks that involve professional judg-ments in environments with relatively little structure where auditorsusually make assessments with a great level of cognitive effort. Sincethe case scenario in the current study is quite structured and limitedto specific choices, the extent to which framing effects still prevail inunstructured accounting/auditing assessments/judgments remainsunanswered. If framing effects do indeed exist in these contexts, thensteps need to be taken soon to improve the quality of the decisions bydeveloping remedial approaches and techniques. Kühberger (1995) hassuggested that providing complete information can alleviate the impactof framing effects. Future studies should focus on how to implementsuch an approach at a reasonable cost and on learning how accountinginformation can be presented to yield higher quality decisions.
An Empirical Examination of Framing Effects 61
REFERENCESArkes, H. 1991. Costs and benefits of judgment errors. Psychological
Bulletin 110: 486–498.Arnold, V. 1997. Judgment and decision making, Part I; The impact of
environmental factors. In Behavioral Accounting Research: Founda-tions and Frontiers, edited by V. Arnold, and S. G. Sutton. Sarasota,FL: American Accounting Association.
Ashcraft, M. H., and J. Battaglia. 1978. Cognitive arithmetic: Evidencefor retrieval and decision processes in mental arithmetic. Journal ofExperimental Psychology: Human Learning and Memory 4: 527–538.
Ashton, R. H., and A. H. Ashton. 1995. Perspectives on judgment anddecision-making research in accounting and auditing. In Judgmentand Decision-Making Research in Accounting and Auditing, edited byR. H. Ashton, and A. H. Ashton. New York, NY: Cambridge UniversityPress.
Barton, S. L., D. D. Duchon, and K. J. Dunegan. 1989. An empiricaltest of Staw and Ross’s prescriptions for the management of escala-tion of commitment behavior in organizations. Decision Sciences 20:532–544.
Boritz, J. E. 1997. Debiasing framing effects in auditors’ internal con-trol judgment and testing decisions. Contemporary Accounting Re-search 14: 79–88.
Bruns, W., Jr., and S. McKinnon. 1993. Information and managers: Afield study. Journal of Management Accounting Research 5: 84–108.
Chow, C. W., P. Harrison, T. Lindquist, and A. Wu. 1997. Escalatingcommitment to unprofitable projects: Replication and cross-culturalextension. Management Accounting Research 8: 347–361.
Davidson, D. 1995. The representativeness heuristic and the conjunc-tion fallacy effect in children’s decision making. Merrill-Palmer Quar-terly 41: 328–346.
Elffers, H., and D. J. Hessing. 1997. Influencing the prospects of taxevasion. Journal of Economic Psychology 18: 289–304.
Emby, C. 1994. Framing and presentation mode effects in professionaljudgment: Auditors’ internal control judgments and substantive test-ing decisions. AUDITING: A Journal of Practice & Theory 13: 102–115.
———, and D. Finley. 1997. Debiasing framing effects in auditors’ in-ternal control judgments and testing decisions. Contemporary Account-ing Research 14: 55–57.
Fagley, N. S. 1993. A note concerning reflection effects versus framingeffects. Psychological Bulletin 113: 451–452.
———, and D. M. Miller. 1997. Framing effects and arenas of choice:Your money or your life? Organizational Behavior and Human Deci-sion Processes 71: 355–373.
Friedman, M., and L. J. Savage. 1948. The utility analysis of choicesinvolving risks. Journal of Political Economy 56: 279–304.
Frisch, D. 1993. Reasons for framing effects. Organizational Behaviorand Human Decision Processes 54: 399–429.
Gelman, R. 1972. The nature and development of early number con-cepts. In Advances in Child Development and Behavior, edited by H.W. Reese, and L. P. Lipsitt. New York, NY: Academic Press.
Gigerenzer, G., U. Hoffrage, and H. Kleinbölting. 1991. Probabilisticmental models: A Brunswikian theory of confidence. PsychologicalReview 98: 506–528.
62 Chang, Yen, and Duh
Hammond, J. S., R. L. Keeney, and H. Raiffa. 1998. The hidden traps indecision making. Harvard Business Review (September-October): 47–58.
Healey, P. M., and K. G. Palepu. 1993. The effects of firms’ financial dis-closure strategies on stock price. Accounting Horizons (March): 1–11.
Jackson, B. R., and M. W. Spicer. 1986. An investigation of under- oroverwitholding of taxes on taxpayer compliance. Working paper, Uni-versity of Colorado.
Jamal K., P. E. Johnson, and R. G. Berryman. 1995. Detecting framingeffects in financial statements. Contemporary Accounting Research 12:85–105.
Johnson, P. E., K. Jamal, and R. G. Berryman. 1991. Effects of framingon auditor decisions. Organizational Behavior and Human DecisionProcesses 50: 75–105.
Kahneman, D., and A. Tversky. 1979. Prospect theory: An analysis ofdecisions under risk. Econometrica 47: 263–291.
Kennedy, J. 1993. Debiasing audit judgment with accountability: Aframework and experimental results. Journal of Accounting Research31: 231–245.
———. 1995. Debiasing the curse of knowledge in audit judgment. TheAccounting Review 70: 249–273.
Kirk, R. E. 1995. Experimental Design: Procedures for the Behavioral Sci-ence. Third edition. Pacific Grove, CA: Brooks/Cole PublishingCompany.
Klaczynski, P. A., and J. Fauth. 1997. Developmental differences inmemory-based intrusions and self-serving statistical reasoning bi-ases. Merrill-Palmer Quarterly 43: 539–566.
———, and G. Narasimham. 1998. Representations as mediators of ado-lescent deductive reasoning. Developmental Psychology 34: 865–881.
Kogan, N., and M. A. Wallach. 1964. Risk-Taking: A Study in Cognitionand Personality. New York, NY: Holt, Rinehart, and Winston.
Kühberger, A. 1995. The framing of decisions: A new look at old prob-lems. Organizational Behavior and Human Decision Processes 62: 230–240.
———. 1998. The influence of framing on risky decisions: A meta-analy-sis. Organizational Behavior and Human Decision Processes 75: 23–55.
Levin, I. P., S. L. Schneider, and G. J. Gaeth. 1998. All frames are notcreated equal: A typology and critical analysis of framing effects. Or-ganizational Behavior and Human Decision Processes 76: 149–188.
Li, S. 1998. Can the conditions governing the framing effect be deter-mined? Journal of Economic Psychology 19: 133–153.
Libby, R., and J. Luft. 1993. Determinants of judgment performance inaccounting settings: Ability, knowledge, motivation, and environment.Accounting, Organizations and Society 5 (18): 425–450.
———. 1995. The role of knowledge and memory in audit judgment. InJudgment and Decision-making Research in Accounting and Auditing,edited by R. H. Ashton, and A. H. Ashton. New York, NY: CambridgeUniversity Press.
Luft, J. 1994. Bonus and penalty incentives contract choice by employ-ees. Journal of Accounting & Economics 18: 181–206.
Martinez-Vazquez, J., G. B. Harwood, and E. R. Larkins. 1992. With-holding position and income tax compliance: Some experimental evi-dence. Public Finance Quarterly 20: 152–174.
An Empirical Examination of Framing Effects 63
McMillan, J. J., and R. A. White. 1993. Auditors’ belief revisions andevidence search: The effect of hypothesis frame, confirmation bias,and professional skepticism. The Accounting Review 68: 443–465.
Newberry, K. J., M. J. Reckers, and R. W. Wyndelts. 1993. An examina-tion of tax practitioner decisions: The role of preparer sanctions andframing effects associated with client condition. Journal of EconomicPsychology 14: 439–452.
O’Clock, P., and K. Devine. 1995. An investigation of framing and firmsize on the auditor’s going concern decision. Accounting and Busi-ness Research 25: 197–207.
Pany, K., and P. M. J. Reckers. 1987. Within- vs. between-subject ex-perimental designs: A study of demand effects. AUDITING: A Journalof Theory & Practice 7 (1): 39–53.
Reyna, V. F., and C. J. Brainerd. 1990. Fuzzy processing in transitivitydevelopment. Annual of Operations Research 23: 37–63.
———, and ———. 1991a. Fuzzy-trace theory and framing effects inchoice: Gist extraction, truncation, and conversion. Journal of Be-havioral Decision Making 4: 249–262.
———, and ———. 1991b. Fuzzy-trace theory and children’s acquistionof mathematical and scientific concepts. Leaning and Individual Dif-ferences 3: 27–59.
———, and S. C. Ellis. 1994. Fuzzy-trace theory and framing effects inchildren’s risky decision making. Psychological Science 5: 275–279.
———, and C. J. Brainerd. 1995. Fuzzy-trace theory: An interim syn-thesis. Learning and Individual Differences 7: 1–75.
Rutledge, R. W., and A. Harrell. 1993. Escalating commitment to anongoing project: The effects of responsibility and framing of account-ing information. International Journal of Management 10: 300–313.
———, and ———. 1994. The impact of responsibility and framing ofbudgetary information on group-shift. Behavioral Research in Account-ing 6: 92–109.
———. 1995. The ability to moderate recency effects through framing ofmanagement accounting information. Journal of Managerial IssuesVII: 27–40.
Sanders, D. L., and R. W. Wyndelts. 1989. An examination of taxpractioners’ decisions under uncertainty. Advances in Taxation 2: 41–72.
Schadewald, M. S. 1989. Reference point effects in taxpayer decisionmaking. The Journal of the American Taxation Association (Spring):68–84.
Schepanski, A., and D. Kelsey. 1990. Testing for framing effects in tax-payer compliance decision. The Journal of the American Taxation As-sociation Fall: 60–77.
———, and T. Shearer. 1995. A prospect theory account of the incometax withholding phenomenon. Organizational Behavior and HumanDecision Processes 63: 174–186.
Schneider, S. L. 1992. Framing and conflict: Aspiration level contin-gency, the status quo, and current theories of risky choice. Journalof Experimental Psychology: Learning, Memory, and Cognition 18: 1040–1057.
Sharp, D, J., and S. B. Salter. 1997. Project escalation and sunk costs:A test of the international generalizability of agency and prospect theo-ries. Journal of International Business Studies 28: 101–121.
64 Chang, Yen, and Duh
Simon, H. 1956. Rational choice and the structure of the environment.Psychological Review 63: 129–138.
Stone, E. R., F. Yates, and A. M. Parker. 1994. Risk communication:Absolute versus relative expressions of low-probability risks. Organi-zational Behavior and Human Decision Processes 60: 387–408.
Tversky, A., and D. Kahneman. 1981. The framing of decisions and thepsychology of choice. Science 211: 453–458.
van Schie, E. C. M., and J. van der Pligt. 1995. Influencing risk prefer-ence in decision making: The effects of framing and salience. Organi-zational Behavior and Human Decision Processes 63: 264–275.
Wang, X. T., and V. S. Johnston. 1995. Perceived social context andrisk preference: A reexamination of framing effects in a life-death de-cision problem. Journal of Behavioral Decision Making 8: 279–293.
———. 1996. Framing effects: Dynamics and task domain. Organiza-tional Behavior and Human Decision Processes 68: 145–157.
Wolfe, C. 1995. Information seeking on Bayesian conditional probabil-ity problems: A fuzzy-trace theory account. Journal of Behavioral De-cision Making 8: 85–108.
Yaniv, G. 2000. Tax compliance and advance tax payments: A prospecttheory analysis. National Tax Journal LII: 753–764.
Zickar, M. J., and S. Highhouse. 1998. Looking closer at the effects offraming on risky choice: An item response theory analysis. Organiza-tional Behavior and Human Decision Processes 75: 75–91.